How CPG Brands Choose an Amazon Agency

Selling a consumer packaged good on Amazon is a repeat-purchase business dressed as a search-advertising one. The first order is an acquisition cost rather than a transaction, and nearly every Amazon decision follows from that: how aggressively you bid, whether you build multipacks, whether you enroll in Subscribe & Save, and which number tells you advertising is working. This page starts with how to choose whoever runs the channel, because that is the question most brands arrive with.

How to choose an Amazon agency

There are four common ways to get Amazon run, each genuinely better than the others in specific circumstances.

In-house. You hire someone whose job is Amazon. This is right when Amazon is a dominant share of revenue and the volume justifies a salary plus the tooling, design, and copy support that person needs. It is wrong when Amazon is a slice of three people's jobs, which is the most common way the channel stalls: nobody owns it, so it gets the leftover hours.

A freelancer or contractor. Fast to start, inexpensive, often very good at one thing. The failure mode is coverage. Amazon problems do not arrive on schedule, and one contractor with several clients cannot absorb a suppressed listing, a stranded-inventory event, and a Q4 capacity limit in the same week. Ask what happens when they take a vacation.

An ad-only agency or point solution. Someone runs the campaigns and nothing else. This works when listings, inventory, and account health are already in good shape and advertising is the only gap. It fails when the ads are fine and the problem is upstream. Sending paid traffic to a page that does not convert wastes the click and depresses the conversion rate on the listing. Fixing the page is cheaper than buying more traffic to it.

Full account management. One team owns advertising, content, catalog, inventory planning, and the account-health work nobody enjoys. The advantage is that finding the problem and fixing it live with the same people, so "your problem is the main image, not the bid" is something they act on instead of assigning back to you. The disadvantage is that it costs more than a point solution and requires you to delegate. If you want to approve every keyword, this model wastes your money. That is the model we run; what it covers is written up separately.

Questions worth asking on a first call

• Who will touch my account each week, and do I meet them before I sign? The common arrangement is a sales team that wins the account and a delivery team you meet afterward. Ask whether the person assessing your account is the person who will run it.
• Show me a report you would send in month two. Not a sample deck, a real one. If it is a screenshot of the advertising console with no reference to total sales, margin, or inventory, you have learned something.
• How do you set ad budget? A fixed monthly number caps you exactly when something starts working.
• What would you tell us not to do? A partner who has never recommended against their own scope is an order-taker. That is an expensive pair of hands, not an advisor.
• What is the out-clause, in days, and what does pricing look like in writing? Both answers should arrive without a second meeting scheduled to deliver them.
• What happens when we go out of stock? Watch whether they treat this as an operations question or as your problem. Inventory is growth work, not back office.

Warning signs

A revenue promise made before anyone has opened your account is the biggest one. Hitting a sales number is easy if you are allowed to waste money doing it, and nobody who has not seen your data can forecast it honestly. Related: proposals built entirely on ROAS, with no mention of total sales, margin, or organic mix.

Then the structural ones. Long out-clauses exist to exploit inertia rather than to earn renewal. Branded and generic keywords sharing a campaign is a tell that the account is run from a playbook imported off another platform. Reporting that only shows what went up. And an unwillingness to name which brands they are not a fit for.

Why CPG advertising on Amazon works differently

Amazon is a buyer-intent platform. The person typing "organic pasta sauce" is holding a wallet, not browsing, which is why playbooks built to create demand travel badly here. Four structural facts change how a CPG brand should target.

Low unit price compresses your tolerance for a bad click. A single jar carries a small contribution margin, so a cost per click that is trivial in durable goods is fatal in a pantry category. That does not mean bid low; it means bid precisely, and treat broad match as discovery rather than strategy.

Repeat purchase makes the first order an acquisition. If buyers come back without you paying again, the honest question is what it costs to acquire a customer, not an order. Brands that ignore this underbid on high-intent terms and hand the customer to whoever was patient.

Pantry-loading changes what you should advertise. Shoppers buying consumables buy in quantity, and a multipack carries more contribution per order and absorbs fulfillment fees far better than a single unit. If your catalog only offers singles, your advertising is fighting your merchandising.

Seasonality is not a promotion calendar. Shelf-stable demand curves track weather, holidays, and category buying habits, and short-dated products add a harder constraint: inventory you cannot sell through in time is a write-off, not a carryover.

Keyword and product targeting for CPG

Assume your category terms are expensive and crowded, and that the money is in the modifiers. "Hot sauce" is a bidding war. "Low sodium hot sauce" and the dietary claim you actually own are where a smaller brand converts. Automatic campaigns find those terms and are more controllable than their reputation suggests: you can set a separate bid per targeting group, close match, loose match, substitutes, and complements. Manual campaigns own the winners at a bid you set per keyword.

Keep branded terms in their own campaigns, always. They convert at a high rate and drag reported efficiency upward, which makes an unprofitable generic term in the same campaign look fine for months.

Product targeting, meaning specific ASINs rather than search terms, is underused in CPG:

• Defensive targeting. Buy placements on your own detail pages and inside your own variation family. If you do not occupy that space, a competitor will, at the moment a shopper has decided to buy your category.
• Conquest targeting. Buy placements on comparable products where you have a defensible reason to win the click: a certification, a cleaner panel, a better price per ounce, or a pack size they do not offer. Without one, you are paying to lose.

Sponsored Products, Sponsored Brands, and Sponsored Display

Sponsored Products is the workhorse and where most CPG spend belongs. It puts one item in front of a search with purchase intent, and the sales it drives count toward the same unit velocity and Best Sellers Rank that organic placement tracks. Amazon has never published how its search ranking treats a sale by acquisition source, so anyone describing that mechanic with confidence is guessing. What is observable is that a sponsored unit and an organic unit are both a unit sold.

Sponsored Brands sells the brand rather than the SKU, which matters for CPG because your assortment is the pitch. It does the most work on your own branded search, where the goal is to own the top of the page instead of letting a competitor rent part of it.

Sponsored Display covers retargeting and audiences, including shoppers who viewed and did not buy. A consumable has natural reorder windows to target against.

One prerequisite sits under most of this page. Sponsored Brands and Sponsored Display require Brand Registry, as do A+ content, a Brand Store, Vine, and Brand Analytics. Brands whose listings were created by distributors are the most likely to need all six and the least likely to have registered. Settle that first.

Reading TACoS against ACoS when customers reorder

ACoS divides ad spend by the sales credited to that spend inside an attribution window, 7 days for Sponsored Products on the seller side and 14 days on the vendor side, so a campaign spending $200 that returns $1,000 in attributed sales runs a 20% ACoS.

That window is the whole problem for a consumable. A shopper who buys a jar in March and reorders in May produced one attributed sale and one uncredited one, and no amount of campaign optimization makes ACoS show you the second. In a category whose economics are built on the reorder, the metric is blind to the part that pays. TACoS divides ad spend by total Amazon sales, ad-attributed and organic together:

TACoS = Ad Spend / Total Amazon Sales * 100

TACoS trending down while total sales grow is the signature of advertising that is compounding rather than renting. TACoS flat or rising while sales stay flat means you are buying the same revenue at an increasing price. We wrote a longer breakdown of where the two disagree in TACoS vs. ACoS, including the scenarios where each gets the read backwards. One rule sits underneath both: budget as a percentage of sales, not a fixed monthly cap. A campaign that exhausts its budget by noon sent the cheap afternoon clicks to somebody else.

Amazon operations for CPG brands

Advertising gets the attention, but operations is where the most avoidable money goes, because it is invisible until it breaks.

Catalog and listing hygiene. One product, one detail page. Duplicate ASINs split your reviews and your rank, and they are common where listings were first created by distributors. Get the browse node right; it decides which category you rank in and which filters you appear under. Keep GTINs consistent, because the barcode decides whether an item matches your existing detail page or spawns a second one beside it.

Variations and multipacks. A variation family is built deliberately, not inferred from your barcodes: you create a parent, declare a theme such as size, flavor, or count, and attach the children by relationship. Done right it concentrates review volume and lets a shopper switch flavor or size without leaving the page. Pack architecture sets your fee exposure, your entry price, and whether you have anything to sell to a shopper who already likes the product. If the assortment is wrong, better bidding will not save it.

Inventory and replenishment. An Amazon listing behaves like a freight train. Momentum takes months to build and one stock-out stops it cold, taking rank, review velocity, and ad efficiency with it. Forecast against lead time plus inbound transit plus receiving time, not last month's sales. Check your FBA capacity limits before a seasonal peak rather than during it: they are set monthly, in cubic feet, at the account level, with a capacity manager for requesting more space. Amazon retired the older ASIN-level restock limits in 2023, so a plan built around per-ASIN caps is a plan against a system that no longer exists. Treat aged inventory as an active cost, because storage fees escalate on the units that are not selling.

FBA versus seller-fulfilled. For most shelf-stable CPG, FBA wins on the math and on speed, and the Prime badge materially affects conversion. The badge is not exclusive to FBA: Seller Fulfilled Prime carries it too, in exchange for nationwide delivery speed standards, on-time shipment thresholds, buying shipping through Amazon, a trial period, and a program fee. That exception matters most where FBA is weakest: oversized items punished by dimensional weight, short-dated or temperature-sensitive goods, and products with fulfillment restrictions. Compare only the line items that differ: FBA fulfillment fees, monthly and aged storage, inbound placement, and returns processing on one side, against your own warehousing, pick and pack labor, and outbound freight on the other.

Vendor Central versus Seller Central. A structural choice, not a preference. Selling to Amazon as a vendor (1P) means purchase orders, wholesale margins, and Amazon controlling retail price and much of the content. Selling as a third party (3P) means you keep price and content control and carry fulfillment and service. What neither gives you is the customer: Amazon owns that relationship in both. FBA orders return anonymized buyer information, and a merchant-fulfilled shipping address may be used to ship that order and nothing else. What 3P does give you is cohort-level visibility through Brand Analytics, repeat-purchase and Subscribe & Save behavior, and permissioned messaging to brand followers and past customers. For most growing CPG brands 3P is the better default, because price and content control are the leverage you need. Hybrids are harder to run than they look.

Chargebacks and compliance. On the vendor side, chargebacks for late shipment notices, labeling errors, and routing mistakes accumulate quietly and are frequently disputable. On the seller side, the equivalent drains are stranded inventory, suppressed listings, restricted claims language in food and personal care, and reimbursement claims that expire before anyone files them. Amazon now auto-reimburses inventory lost inside a fulfillment center and values lost or damaged FBA units at manufacturing cost rather than estimated sale price, so the old 18-month backlog of unfiled claims is gone. What is left sits in inbound shipment discrepancies, removals, and fee or dimension misclassification, and the windows are short: roughly 60 days for fulfillment-center loss claims, with separate and shorter windows for inbound and removal discrepancies. The risk is a window closing while nobody is watching.

Retail readiness and discovery

Retail readiness is the discipline of never sending a paid click to a page that cannot close it. It is upstream of everything above.

Content that converts. Judge every asset on a phone, mid-scroll, in a few seconds, because that is how the purchase happens. Main image on white, product legible at thumbnail size. Use the secondary images to say what a bullet never gets read for: what is in it, what is not, how big it is, and what makes it different. A differentiator you paid for belongs in image two rather than the fifth bullet. A+ content carries the brand story and the comparison chart.

Reviews. Volume and recency drive conversion, and for a new CPG SKU the cold-start problem is the whole battle. Whether recency moves organic rank is inferred rather than documented; the conversion effect alone is reason enough to plan around it. Amazon Vine is the sanctioned way to seed early reviews, and it is narrower than most brands expect: it requires Brand Registry, is limited to 30 units per parent ASIN, is open only to ASINs with fewer than 30 existing reviews, and carries a per-ASIN enrollment fee above the free tier. For a brand launching a dozen SKUs, those limits are the planning problem. Everything else runs through product quality, a fast reorder loop, and not doing the things that get an account suspended, starting with review solicitations in packaging inserts.

Brand store. Your store is where Sponsored Brands traffic should land when the pitch is the assortment rather than one SKU, and where a shopper who found you in a grocery aisle ends up. Build it as a storefront window, not a directory.

None of this finishes. Listings drift and a variation family breaks the week you stop looking at it, and whoever owns the channel owns that maintenance; how the ongoing work is structured is written up separately.

Terms worth being precise about

What is PCOGS on Amazon? Product cost of goods sold, a Vendor Central metric. It is the vendor cost of the units Amazon sold to customers in a period: sell-through valued at your wholesale cost rather than Amazon's retail price, and not the same thing as what Amazon paid you on purchase orders. Those two populations of units can sit months apart, which is how a Q4 buy-in gets paid in October and booked as PCOGS in November and December. Vendor allowances and accruals (damage allowance, marketing development funds, freight) are calculated against it, which is why it governs negotiations.

ACoS and TACoS. Advertising cost of sales divides ad spend by the sales attributed to it inside the attribution window. Total advertising cost of sales divides ad spend by all Amazon sales, organic included. The first measures a campaign, the second a business.

1P and 3P. First party means you sell to Amazon and Amazon resells to the customer. Third party means you sell directly to the customer on Amazon's marketplace.